Alarmed by Tesla’s self-driving test, state legislators demand answers from DMV
latimes.com
latimes.com
1. The paid software unlock for “Full Self Driving” (currently in “beta”) available to everyone, which provides automatic driving from on-ramp to off-ramp, and a handful of parking features like auto-park and summon: https://www.tesla.com/support/autopilot#usingautopilot
2. The invite-only “FSD Beta” that unlocks the above FSD features on city streets, which is available to a subset of customers w/#1 who have very high Tesla Safety Scores.
USB 2.0 got away with Full Speed/High Speed (with high being more than full)
Actually, NHTSA defines[0] level 5 as: "An automated driving system (ADS) on the vehicle can do all the driving in all circumstances. The human occupants are just passengers and need never be involved in driving."
[0]: https://www.nhtsa.gov/technology-innovation/automated-vehicl...
That's correct; "Full speed" refers to the higher speed profile for USB1.1. It's a historical artifact, at this point, and wasn't generally used for marketing purposes.
Wait, does this mean people are actually paying extra money to become test dummies for this?
“No, you should do some basic research first.”
The FSD beta is no additional cost over the (currently) $10,000 package that adds additional AP capabilities and hardware upgrades. My computer was already upgraded for example.
Why should I? I read the article and can ask/talk about it here--discussion is literally the point.
You can go through the order flow on Teslas site and see this quickly.
i’m one of them AMA
https://twitter.com/mostlyharmlessz/status/14694217625553510...
The technology being tested seems half baked and not yet ready for testing on a public roads. At the very least, their approach to software updates does not seem like its being sufficiently tested before deployment. This video[1] linked in another thread is an example of what I mean.
[1] https://twitter.com/mostlyharmlessz/status/14694217625553510...
Sorry, but they same system lets the driver play video games while driving down the road[1] which is kinda the opposite of what you claim.
1 - https://www.nytimes.com/2021/12/07/business/tesla-video-game...
What are the arguments for why non-owners should support this approach?
What are the arguments for why non-owners should support this approach?
This beta adds the same benefits and more on city streets.
If I could choose to have only FSD teslas around me on the highway, or only human drivers, I'd pick the FSD every time.
There are thousands of clips of Tesla's Fatal Self Driving being far too aggressive for American roads (https://nitter.net/taylorogan/status/1469404579439824899).
Meanwhile, companies like AutoX have the problem of being far too conservative for Asian roads (https://www.youtube.com/watch?app=desktop&v=TFEvkmvIjVo). AutoX has some of the best edge case handling I've seen.
I remember the first time I drove as a passenger through Manhattan with my father driving, and couldn't get my head around his change in driving style. Was a side of him I'd never seen before - driving aggressively like a cab driver. I'm sure it came from his years of living in NYC.
It's one of the things that makes me think that true self-driving may be in the AGI bucket.
Genuine question: if this happens, will it kill Tesla to refund all of this money? I don't pay close enough attention to their financials to know myself. I assume (and hope) not.
it is also not that expensive to retrofit most new hardware, especially if it is just new silicon.
As an aside, I’ve driven maybe 50,000 miles in my lifetime. Of those, 47,000+ are using pre-2005 vehicles with zero “advanced safety features”. However, I can recall two times in my driving career where I felt collision was imminent, and both of them occurred in the <5% of the time I was driving with “advanced safety features”. Yes, the computer was able to step in and avoid a collision, but it strikes me as odd that the only times I’ve needed it are when it’s available. In my opinion these new cars lull me into a false sense of security where I do not give appropriate attention to road conditions. Has anyone noticed the same? I fear that as the technology gets better and better people will pay less and less attention, putting more and more lives at risk when things inevitably go wrong.
As for the “lightweight Teslas”, I looked it up and in fact S and X’s are heavier than my Big Truck. Though 3’s are indeed lighter.
I don’t understand rich guys who fork over for what is essentially a baby carriage pushed by a robot nanny. Where is your self-respect?
I once owned a sports car. I bought it because I enjoy driving. I am saying that I, as a red blooded American man, do not understand why other men who are in the market for an expensive car would choose to purchase what is functionally a nannyless pram.
As you can see with me understanding exactly what you said, you tend to jump to conclusions while not understanding things. Maybe that explains the inability to grasp "rich people" motives.
We should treat the AIs just as children at this point. They are simply learning.
I’m not sure this would have helped the Tesla that steered into oncoming traffic. Luckily for the other driver the Tesla driver reacted and crashed off on the other direction. If he hadn’t done that I don’t think the other driver would have had the time to react even if he was aware of the danger the Tesla posed.
While you can dispute the data, it’s not as simple as “car makes one type of mistake”, “feature should be banned”.
Moreover, Tesla influencers on social media tend to cherry pick and show only the nicer parts of their FSD Beta drives where no incidents and disengagements occur. Whereas you will often see in the comments some Beta users disagreeing and often showing the scary bits. And of course the pattern of Tesla influencers acting like an echo chamber absolutely disagreeing with anything critical of Tesla, jumping to Musk's defense at every turn because..."My mission!". An example:
https://mobile.twitter.com/taylorogan/status/146940457943982...
A thread of the scariness of this Beta test on public roads:
https://twitter.com/Tweet_Removed/status/1438553523369615364...
Elon Musk has a history of overestimating Tesla's capabilities when it comes to autonomy. And also flouting regulations to reach the end goal. He wants to make people believe in the illusion that Tesla are in some unsurmountable lead when it comes to autonomy but they are actually playing catchup and are dead last because of the approach they took.
Here's a timeline of Musk dangling the carrot of full autonomy (SAE Level 5) for over 6 years now.
https://www.reddit.com/r/SelfDrivingCars/comments/n6nsmt/elo...
But people argue, "Oh, but they are getting better and eventually will iron out all the edge cases!". This is not a matter of just making some settings change and things magically fix themselves. These are safety critical systems and not considering edge cases while advertising the feature as "Full Self-Driving" can lead to fatalities. The edge cases don't seem to be finite and no amount of "Dojo will solve this!", "The system is learning more and more and will reach there!" will solve the problem with their approach. At some point, he will have to eat humble pie and incorporate more sensors to increase safety. But with the ongoing chip shortage and the fact that they have to sell as many cars as possible lest the stock promotion crumbles makes it an even harder task.
It's always some next version that will blow your mind. It's always the next version that's almost there.
https://twitter.com/elonmusk/status/1435967157662150675?s=20
https://twitter.com/elonmusk/status/1469541536128020485?s=20
Damn, the part where he tried to take over and the car wouldn't let him for a while, was downright scary
Reminds me of a famous quote...
> "I'm sorry, Dave. I'm afraid I can't do that."
I fell like there should be a hardware way of taking over that overwrites anything the software does.
I also believe the driver in that video doesn't have the level-headedness to be testing anything anywhere. Which is a criticism both of the driver and of Tesla for letting people like that run the software on the open road.
And it wasn't even rush hour through some crowded European city like Rome with people on scooters zooming in every directions, in poor weather conditions at night, it was broad daylight in a city with barely any traffic.
If it can't handle basic city driving properly, then I'm sorry.
Asking as someone with no first hand experience with these systems - do they complete 99% of trips with no disengagements or unexpected behavior, 99% of time driving, something else?
Not defending Tesla for all of those mistakes, but as a human, it’s easy to end up going down the wrong lane.
The "edge" cases are pretty much everything else. There are a near infinite number of possibilities here and it is clear that no one is anywhere near close enough to covering them all. Stopping for a pedestrian when turning right is not some completely unforeseeable event.
Human driver as in: average driver, or best driver?
The burden of proof is on the person making the claim of safety. For Autopilot, the argument was made comparing known AP deaths per total AP miles vs auto deaths vs. driving miles of the entire US fleet... when the former almost certainly didn't include a share of the deaths and was driven under less demanding conditions (highway miles).
Ergo there's a predictable and preventable source of accidents that must be suppressed.
a) [Broadly speaking], cars [from traditional big autos] with ADAS features are objectively safer, and,
b) Tesla Autopilot is [in traditional sense] an ADAS system, which,
c) [implies this set of facts alone make Tesla cars and implementations statistically safer] is cool, huh? Would be silly if you didn’t agree with me.
I don’t call it clever.
The utilitarian argument for driverless cars needs to be that driverless cars are better at handling the edge cases than humans, because that's when accidents happen.
The average situation is maybe 0.99 accident-free (made up number for illustration). You improve on that by being 0.995 accident-free, which is better.
I agree that “safer than some human drivers in some conditions” is insufficient.
Maybe I'm confusing the idea of average and mode, so if I am an example of what you have in mind for an average situation would help.
https://www.theverge.com/platform/amp/2021/10/21/22738834/te...
The FUD is _insane_ with Tesla.
Latest:
Improved object detection network architecture for non-VRUs (e.g. cars, trucks, buses). 7% higher recall. 16% lower depth error, and 21% lower velocity error for crossing vehicles.
New visibility network with 18.5% less mean relative error.
New general static object network with 17% precision improvements in high curvature and nighttime cases.
Improved stopping position at unprotected left turns while yielding to oncoming objects, using object predictions beyond the crossing point.
Allow more room for longitudinal alignment during merges by incorporating modeling of merge region end.
Improved comfort when offsetting for objects that are cutting out of your lane.
Those numbers don't look useful for judging safety. How would one compare those against anything else?
If a tesla is at-fault, there’s another non-tesla party in the accident and police and dmv usually involved.
We can't really draw any conclusions from self-selected video clips.
I do think the data should be openly available for all to study an analyze. If that helps competitors, then that's the price for testing in public.
Latest example: https://twitter.com/mostlyharmlessz/status/14694217625553510...
Even if what you’re saying is right about influencers picking and choosing the best videos, it’s clear that there are videos of poor behaviour as well.
With so many videos out there I think it’s pretty clear what you’re getting with FSD regardless of the name. It is glitchy and unreliable and can get random safety issues with software regressions.
At the same time when it works it works pretty well, and it does also seem to be improving rapidly based on the videos that I’ve seen. A long way to go but it’s getting better.
It’s also not clear they’re dead last - they may be but there’s lots of people using it now and showing videos, which is harder to see for other companies.
You might have good points here but I don’t think it’s as black and white as you make out regarding how effective Tesla’s approach is with respect to safety, degree of progress, feasibility with current sensors, feasibility in general, misleading market, and general motivations and morals.
> He wants to make people believe in the illusion that Tesla are in some unsurmountable lead when it comes to autonomy but they are actually playing catchup and are dead last because of the approach they took.
I'm sorry, but yeah, the results in that video are worse than what we had 14 years ago after the DARPA Urban Challenge. I'm actually astounded how bad it is. An insurmountable lead this is not, and points to how much of an advantage LIDAR is.
Also, while we all have our robot failure blooper reels, this is a system that's live and in production, and Tesla should be ashamed to release this even under the "Beta" label.
I belive Teslas approach might currently feel behind but has much more “headroom”. As Elon states the team have gone down many paths that lead to a local maximum which is short of the actual end goal. They firmly belive that an end-to-end system has to be based on vision and has to be able to drive in any situation the very first time without localizing itself to a preset path.
Moreover, what I said was after the DUC. The results you see at the DUC were the culmination of a little over a year’s worth of work. They were at best prototypes constructed under great pressure in a competition setting. A lot of what you see there was held together by duct tape and 80/20, and the code was not something you’d want to bet your life on.
But things did improve rapidly soon after, as results were disseminated, the community began to understand what worked and what didn’t in designing these systems.